Amazon has a new kind of customer it does not want: one that is not human.
In early September, Meta launched Muse, a personal AI agent that can send emails, complete forms and book travel. With Link’s wallet for agents, provided by Stripe, it can also make purchases. The basic version is free, and it quickly topped the U.S. free-app charts.
Amazon promptly banned Muse agents from its site. The company argued that outside applications making purchases for customers should operate openly and respect a retailer’s decision about whether to participate.
Adobe Stock
This was not Amazon’s first battle with AI shoppers. In November 2024, the AI-powered search engine Perplexity launched Buy with Pro, allowing subscribers to purchase a recommended product with one click.
Perplexity’s chief business officer, Dmitry Shevelenko, told Modern Retail that Buy with Pro was “just acting as an agent on behalf of the user that’s looking to make a transaction” and was “complementary” to Amazon’s role as a merchant.
The conflict escalated the following year with Comet, Perplexity’s AI-powered browser. Amazon introduced a technical block, which Perplexity worked around within 24 hours, then issued a cease-and-desist order. When that failed, Amazon sued, arguing that Comet was “covertly” posing as a human customer and violating its terms.
There was also money at stake. Traffic generated by agents complicates how Amazon measures what advertisers are paying for. Advertisers pay for human attention, and Amazon’s advertising business generated $19.8 billion in the second quarter of 2026. In March 2026, Amazon won a preliminary injunction. But in August, a federal appeals court vacated it, ruling that Perplexity was not accessing Amazon’s website. Its customers were, using the agent to help them. Tell an agent to buy something and you are the one buying, not the agent.
Liability Gap
This is partly a fight over advertising revenue. It is also a glimpse of a larger problem: what happens when the thing making a commercial decision cannot be held responsible for it?
Every purchase involves two things: information and the assurances behind it. Forrester surveyed nearly 18,000 business buyers and found that 94 percent had used generative AI during their most recent purchase, up from 89 percent the year before. Twice as many called AI their most important research source as named vendor websites, product experts or sales representatives.
Your company may have been shortlisted for deals nobody told you about and rejected from others, based on evidence you did not supply and cannot see. But agents do not provide assurances. An agent has nothing at stake in a transaction. When it gives a false answer, no one has broken their word because no one gave it.
Buyers have worked this out. Gartner surveyed 645 business buyers and found that 67 percent preferred making a purchase without a sales representative. Yet 69 percent turned to a representative to check what AI had told them. Fifty-one percent said they were more likely to be misled by generative AI, while 49 percent said the same of a salesperson. They trust the machine slightly less than the person paid to sell to them. They use it first anyway.
That is not confusion. It is a division of labor between information and assurance. Forrester found that 13 people inside a company and nine outside it now influence the average business purchase. The machines provide the information. More humans are brought in to carry the risk.
A procurement platform created by GEP shows where the line is being drawn. Its agents can operate across sourcing, supplier management and risk monitoring. But sensitive transactions still require controls and oversight. An agent can help select suppliers and negotiate terms. But the greater the risk, the greater the need for human oversight.
The same pattern appears everywhere. Agents discover, assess and begin the process. Humans approve decisions and step in when something goes wrong.
Amazon can respond by blocking outside agents and building its own. Almost nobody else has that option. If buyers search for a type of product rather than for your company by name, refusing to engage changes only the evidence they use to judge you.
Responding also requires collaboration by parts of a company that rarely work together. Operations holds the information an agent needs. Technology builds the systems it reads. Marketing must reach buyers who are no longer reading the material themselves. Sales manages a relationship the agent has already reordered.
Ask who is responsible for ensuring that an agent reaches the right conclusion about the company, and you will probably receive four polite deflections.
Rewriting Sales
The deeper problem is talent. A study of more than 1,000 top-ranked Wall Street analysts, published in Chasing Stars, found that those who moved firms often performed worse. Their success had depended partly on resources, colleagues and networks they left behind. Those who moved with their teams kept performing. They brought the support system with them.
Machine-led selling turns that problem on its head. Nobody changes firms. Instead, the job changes around them. The work of finding potential customers disappears. The first human conversation happens after the buyer has drawn up a shortlist, rather than helping to create it. The structure that produced outstanding salespeople is dismantled while everyone remains in place. Companies will need new roles, incentives and definitions of good performance. Eventually, buyer agents may negotiate with seller agents while humans approve the outcome, resolve exceptions and protect relationships when an agent gets something badly wrong.
Bank of America is at least approaching the issue in the right order. It has presented the hours saved by introducing AI tools across Merrill and its private bank as time for building deeper relationships with existing clients, rather than simply serving more of them.
Trade between machines will not end human conversation. It will make it rarer and, therefore, more valuable. Amazon’s battle with AI agents is not over. The appeals court ruling was narrow, and it remains unclear how the conflict with Muse will unfold. What is clear is that agents are changing commercial decision-making, who carries the risk and what the humans are still there to do.